US7142732B2 - Unsupervised scene segmentation - Google Patents
Unsupervised scene segmentation Download PDFInfo
- Publication number
- US7142732B2 US7142732B2 US10/312,508 US31250803A US7142732B2 US 7142732 B2 US7142732 B2 US 7142732B2 US 31250803 A US31250803 A US 31250803A US 7142732 B2 US7142732 B2 US 7142732B2
- Authority
- US
- United States
- Prior art keywords
- image
- produce
- top hat
- marker
- interest
- Prior art date
- Legal status (The legal status is an assumption and is not a legal conclusion. Google has not performed a legal analysis and makes no representation as to the accuracy of the status listed.)
- Expired - Fee Related, expires
Links
Images
Classifications
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06T—IMAGE DATA PROCESSING OR GENERATION, IN GENERAL
- G06T7/00—Image analysis
- G06T7/10—Segmentation; Edge detection
- G06T7/11—Region-based segmentation
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06T—IMAGE DATA PROCESSING OR GENERATION, IN GENERAL
- G06T7/00—Image analysis
- G06T7/10—Segmentation; Edge detection
- G06T7/155—Segmentation; Edge detection involving morphological operators
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06V—IMAGE OR VIDEO RECOGNITION OR UNDERSTANDING
- G06V20/00—Scenes; Scene-specific elements
- G06V20/60—Type of objects
- G06V20/69—Microscopic objects, e.g. biological cells or cellular parts
- G06V20/693—Acquisition
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06T—IMAGE DATA PROCESSING OR GENERATION, IN GENERAL
- G06T2207/00—Indexing scheme for image analysis or image enhancement
- G06T2207/10—Image acquisition modality
- G06T2207/10056—Microscopic image
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06T—IMAGE DATA PROCESSING OR GENERATION, IN GENERAL
- G06T2207/00—Indexing scheme for image analysis or image enhancement
- G06T2207/20—Special algorithmic details
- G06T2207/20112—Image segmentation details
- G06T2207/20152—Watershed segmentation
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06T—IMAGE DATA PROCESSING OR GENERATION, IN GENERAL
- G06T2207/00—Indexing scheme for image analysis or image enhancement
- G06T2207/30—Subject of image; Context of image processing
- G06T2207/30004—Biomedical image processing
- G06T2207/30024—Cell structures in vitro; Tissue sections in vitro
Definitions
- This invention relates to an improved method of segmenting objects in an image.
- it relates to a method of transforming an image using Top Hat and Watershed transforms for scene segmentation of an image into regions-of-interest.
- a common task is the selection of objects-of-interest in a captured image.
- a typical task is the selection, or segmentation, of cell nuclei from an image containing many cells. This task is often separated into two sub-tasks.
- the first sub-task is that of scene segmentation.
- the original image is coarsely segmented into regions-of-interest, each containing a single cell-nucleus-of-interest.
- Each region-of-interest may then be viewed as a new, smaller image to process.
- the second sub-task may consist of further segmentation of the cell-nucleus-of-interest within each region-of-interest.
- the second sub-task is the subject of co-pending PCT Patent Application number PCT/AU99/00231 claiming priority from Australian Provisional Patent Application number PP2786 dated 3 Apr. 1998.
- the Patent Application is entitled Method of Unsupervised Cell Nuclei Segmentation and describes a method of active contours using a Viterbi search algorithm for unsupervised segmentation of cell nuclei.
- the first sub-task of scene segmentation operates on an image containing between zero and many objects-of-interest.
- the method of the present invention is for the implementation of scene segmentation.
- the invention resides in an improved method of scene segmentation including the steps of:
- the method may further include the step of delineating a region-of-interest around each segmented object to produce region-of-interest images.
- the method Prior to the Top Hat transformation step, the method preferably includes the step of down-sampling the input image by a factor N to produce a smaller image.
- the method preferably includes the step of up-sampling the eroded and dilated images by a factor of N.
- the up-sampling step suitably uses the same factor of N such that the inner and outer marker images are the same size as that of the input image.
- FIG. 1 is a flow chart showing the steps in the invention
- FIG. 2 shows the input image A before down-sampling and the image A′ after down-sampling
- FIG. 3 shows the treatment of an image as a landscape
- FIG. 4 shows the Jackway Top Hat transform of a signal
- FIG. 5 shows the image T′ which is the result of applying the Jackway Top Hat transform to image A′
- FIG. 6 shows the image I′ which is the result of applying the morphological erosion to image T′
- FIG. 7 shows the image R′ which is the result of applying the binary reconstruction to the image T′ using marker image I′;
- FIG. 8 shows the image M′ which is the result of applying the morphological dilation to the image R′
- FIG. 9 shows the inner marker image I which is the result of applying up-sampling to image I′;
- FIG. 10 shows the image M which is the result of applying up-sampling to image M′
- FIG. 11 shows the outer marker image O which is the result of applying the perimeter operation to image M
- FIG. 12 shows the input image A, with inner marker image I, and outer marker image O, superimposed
- FIG. 13 shows the watershed transform image W of input image A with inner marker image I, and outer marker image O;
- FIG. 14 shows the ROI images which are the result of applying object extraction to image W.
- FIG. 1 shows a flow chart of the steps involved in the scene segmentation process.
- the image produced after each step is indicated by a letter shown in bold.
- the bold letters are referred to throughout the description and may be cross-referenced with FIG. 1 .
- the first step in the segmentation process may be to down-sample an input scene image A to a smaller image A′. This is shown in FIG. 2 . This may be achieved with the use of, for example, quadtree decomposition, although it is envisaged that alternative down-sampling methods could be used.
- N The factor by which the input image is down-sampled, may depend on the size and resolution of the input image.
- the magnification of the input image may also affect the extent to which the input image may be down-sampled. If N takes a value of 2, a 4 pixel by 4 pixel image is down-sampled to 1 pixel. N may take a value of unity, in which case the down-sampling process has no effect on the size of the input image.
- one benefit of down-sampling the input image is the reduction of processing time due to the resultant smaller image containing fewer pixels.
- Another benefit is the reduction of noise, which is smoothed during the down-sampling process.
- contrast in the down-sampled image is better than in the original input image.
- the Jackway Top Hat transform is described in detail in Electronics Letters, Vol 36, No 14, pg 1194–5. Other possible Top Hat transforms are referred to in the Electronics Letters paper.
- the down-sampled gray-scale image A′ (bottom image in FIG. 3 ), is treated as a landscape (top image in FIG. 3 ), in which higher altitude levels represent darker pixels in the gray-scale image and lower altitude levels represent lighter pixels.
- the Top Hat transform works by passing a three-dimensional shape, that of a top hat shape, over a three-dimensional landscape.
- the rim of the top hat is placed at every location of the landscape such that it rests on the highest part of the landscape underneath the top hat.
- the top hat maintains the same attitude, i.e. it cannot rest on two different levels and be at a slant. If a part of the image landscape penetrates the top of the top hat, that part of the image becomes labelled in the transformed image. If the top hat is not penetrated by a particular part of the image, that part is not labelled in the transformed image. This is illustrated in two dimensions in FIG. 4 .
- the result T′ of the Top Hat transform of the down-sampled image A′ of FIG. 3 is shown in FIG. 5 .
- Parts of the landscape that penetrate the top hat are shown in white and those that did not are shown in black.
- a comparison of FIG. 5 with the input cell image in FIG. 2 shows that the location of some of the white areas of the Top Hat image correspond to the locations of objects-of-interest.
- the next step in the scene segmentation process is to remove the white regions shown in FIG. 5 that do not correspond to objects-of-interest. This is achieved by eroding the image of FIG. 5 .
- Morphological erosion is a preferred technique of removing pixels from regions of an image to make the regions uniformly smaller.
- a suitable process for binary morphological erosion by a factor K 1 is shown in the following pseudo-code:
- the result of eroding the image of FIG. 5 is the image I′ shown in FIG. 6 . It can be seen that some of the small white regions of FIG. 5 have been removed completely, whilst the larger white regions have been made smaller. At this stage in the process, the aim is that every white region in FIG. 6 corresponds to a point within an object-of-interest in the input image of FIG. 2 .
- the method of the present invention uses a Watershed transform to obtain a better estimate of the regions-of-interest.
- the Watershed transform requires two inputs. It requires a point within each object-of-interest, an inner marker, and a closed boundary outside each object-of-interest, an outer marker.
- the white regions in FIG. 6 serve as a good inner marker and the generation of a good outer marker, according to the present invention, is described in the following.
- the image I′ of FIG. 6 is used as a starting point for the generation of an outer marker.
- An operation known as binary reconstruction is used to re-grow the white regions T′ of FIG. 5 from the white regions of I′ of FIG. 6 .
- binary reconstruction is used to re-grow the white regions T′ of FIG. 5 from the white regions of I′ of FIG. 6 .
- only the white regions in FIG. 6 are used because these correspond to the objects-of-interest.
- the following pseudo-code demonstrates a suitable binary reconstruction method:
- Block “Binary Reconstruction” Inputs: Image T′, Marker Image I′ Output: Image R′ Algorithm: FOR each Object in Image T′ IF AND(Object K , Marker I′) ⁇ 0 Copy Object K to Image R′ ENDIF RETURN Image R′
- the result of this operation is the image R′ shown in FIG. 7 .
- the reconstructed image R′ of FIG. 7 is then dilated by a factor K 2 according to the following pseudo-code:
- ENDFOR ENDFOR RETURN Image M′ NewImage
- Morphological dilation is a preferred technique to produce the opposite effect of morphological erosion and, in this case, adds pixels to the white regions of FIG. 7 .
- the result M′ of dilation is shown in FIG. 8 .
- the dilation step could occur before binary reconstruction, it is computationally wasteful to do so, since additional calculations are made that are then discarded.
- the final output from the Watershed transform must be at the original input image size and therefore the transform must be performed at the original input image size.
- the inputs for the transform are required to be at the original input image size. Consequently, the dilated image M′ shown in FIG. 8 and the inner marker image I′ shown in FIG. 6 are up-sampled to the original image size.
- the factor by which the images are up-sampled, N is the same as the factor by which the original input image was down-sampled to reduce the original image.
- the inner marker image I is the result of up-sampling I′, and is shown in FIG. 9 .
- the result of up-sampling M′ is the image M, which is shown in FIG. 10 .
- the following pseudo-code exemplifies up-sampling of the image I′ to the image I.
- the same algorithm may be applied to up-sample M′ to M.
- the outer marker required for the Watershed transform may be generated from the up-sampled dilated image by determining perimeters of the up-sampled white regions. The perimeters are shown in the image O in FIG. 11 .
- the result of the perimeterisation algorithm is the image O shown in FIG. 11 .
- the next step in the process is to apply a Watershed algorithm such as is described by Beucher and Meyer in chapter 12, pages 433–482 of “Mathematical Morphology in Image Processing”, published in New York by Marcel Dekker in 1993.
- the result of the Watershed transform using the generated inner and outer markers as inputs, is image W which is shown in FIG. 13 . It can be seen that each object-of-interest has been delineated.
- These images may be used for further segmentation processing, such as for the second segmentation sub-task mentioned above.
Landscapes
- Engineering & Computer Science (AREA)
- Physics & Mathematics (AREA)
- General Physics & Mathematics (AREA)
- Theoretical Computer Science (AREA)
- Computer Vision & Pattern Recognition (AREA)
- Life Sciences & Earth Sciences (AREA)
- Health & Medical Sciences (AREA)
- Biomedical Technology (AREA)
- General Health & Medical Sciences (AREA)
- Molecular Biology (AREA)
- Multimedia (AREA)
- Image Analysis (AREA)
- Image Processing (AREA)
- Apparatus For Radiation Diagnosis (AREA)
Abstract
A method of segmenting objects in an image is described. The method applies a Top Hat algorithm to the image then constructs inner and outer markers for application to the original image in a Watershed algorithm. The inner marker is constructed using binary erosion. The outer marker is constructed using binary dilation and perimeterisation. The method finds particular application for first level segmentation of a cell nucleus prior to detailed analysis.
Description
This invention relates to an improved method of segmenting objects in an image. In particular, although not exclusively, it relates to a method of transforming an image using Top Hat and Watershed transforms for scene segmentation of an image into regions-of-interest.
In computer image analysis, a common task is the selection of objects-of-interest in a captured image. For example, in computer automated cytology, a typical task is the selection, or segmentation, of cell nuclei from an image containing many cells. This task is often separated into two sub-tasks.
The first sub-task is that of scene segmentation. In this sub-task, the original image is coarsely segmented into regions-of-interest, each containing a single cell-nucleus-of-interest. Each region-of-interest may then be viewed as a new, smaller image to process.
The second sub-task may consist of further segmentation of the cell-nucleus-of-interest within each region-of-interest. The second sub-task is the subject of co-pending PCT Patent Application number PCT/AU99/00231 claiming priority from Australian Provisional Patent Application number PP2786 dated 3 Apr. 1998. The Patent Application is entitled Method of Unsupervised Cell Nuclei Segmentation and describes a method of active contours using a Viterbi search algorithm for unsupervised segmentation of cell nuclei.
The first sub-task of scene segmentation operates on an image containing between zero and many objects-of-interest. The method of the present invention is for the implementation of scene segmentation.
Various techniques exist for identifying and segmenting objects in an image. One method of scene segmentation is described by Betel et al [Segmentation and numerical analysis of microcalcifications on mammograms using mathematical morphology; British Journal of Radiology; vol. 70; no. 837; September 1997, pp 903–17], which discloses the use of simple Top Hat and Watershed algorithms of mathematical morphology to automatically detect and segment microcalcifications on digitized mammograms. The Betel process is of limited use in cell nuclei segmentation due to insufficient noise immunity and region-growing constraints.
It is an object of the present invention to provide an improved method of image scene segmentation.
It is a further object of the invention to provide an improved method of segmenting an image scene into regions-of-interest containing cell nuclei for subsequent determination of nuclei characteristics. Other objects will be evident from the following discussion.
In one form, although it need not be the only, or indeed the broadest, form, the invention resides in an improved method of scene segmentation including the steps of:
- (i) transforming an input image by a Top Hat operator;
- (ii) conducting erosion on said Top Hat transformed image to produce an inner marker image;
- (iii) eliminating objects less than an object-of-interest size by reconstructing said Top Hat transformed image with said inner marker image to produce a reconstructed image;
- (iv) conducting dilation on said reconstructed image to produce an outer marker image;
- (v) determining a perimeter of said outer marker image; and
- (vi) transforming said input image using said inner marker image and said perimeterised outer marker image by a Watershed operator to produce a segmented image.
In preference, the method may further include the step of delineating a region-of-interest around each segmented object to produce region-of-interest images.
Prior to the Top Hat transformation step, the method preferably includes the step of down-sampling the input image by a factor N to produce a smaller image.
After said erosion and dilation steps, the method preferably includes the step of up-sampling the eroded and dilated images by a factor of N.
Any suitable value of N may be used. The up-sampling step suitably uses the same factor of N such that the inner and outer marker images are the same size as that of the input image.
Preferred embodiments of the invention will now be described with reference to the following drawings, in which:
The first step in the segmentation process may be to down-sample an input scene image A to a smaller image A′. This is shown in FIG. 2 . This may be achieved with the use of, for example, quadtree decomposition, although it is envisaged that alternative down-sampling methods could be used.
The factor by which the input image is down-sampled, N, may depend on the size and resolution of the input image. The magnification of the input image may also affect the extent to which the input image may be down-sampled. If N takes a value of 2, a 4 pixel by 4 pixel image is down-sampled to 1 pixel. N may take a value of unity, in which case the down-sampling process has no effect on the size of the input image. The method of the present invention may be carried out without changing the size of the input image. Wilson and Spann (Image Segmentation and Uncertainty, Wiley 1988), assuming circular image objects of radius 2r, calculated an optimum value for N as
N=r−1.
N=r−1.
However, one benefit of down-sampling the input image is the reduction of processing time due to the resultant smaller image containing fewer pixels. Another benefit is the reduction of noise, which is smoothed during the down-sampling process. Thus, contrast in the down-sampled image is better than in the original input image.
The following block of pseudo-code achieves a suitable down-sampling:
Block: “Down Sample by factor N” |
Inputs: Image A, factor N |
Output: Image A′ |
Algorithm: |
NewImage = Image A |
FOR |
OldImage = NewImage | |
NewImage = size(X_size(OldImage)/2, Y_size(OldImage)/2) | |
FOR each Pixel in NewImage | |
Pixel NewImageX,Y = |
(OldImage2X,2Y + OldImage2X,2Y+1 + OldImage2X+1,2Y + | |
OldImage2X+1,2Y+1)/4 |
ENDFOR |
ENDFOR |
RETURN Image A′ = NewImage |
The down-sampled image is then subjected to a Top Hat transform, such as the Jackway Top Hat transform,
H=f−((f⊕(B do \B di))⊖B di),
where f is the original image, Bdo and Bdi are outer and inner structuring elements respectively, ⊕, ⊖ and \ are the standard morphological operators of dilation, erosion and set difference respectively. The intermediate image, H, is finally thresholded by a value T to result in the Top Hat image by
Th=1 if H≧T;
Th=0 if H<T
where Th is the Top Hat image. The Jackway Top Hat transform is described in detail in Electronics Letters, Vol 36, No 14, pg 1194–5. Other possible Top Hat transforms are referred to in the Electronics Letters paper.
H=f−((f⊕(B do \B di))⊖B di),
where f is the original image, Bdo and Bdi are outer and inner structuring elements respectively, ⊕, ⊖ and \ are the standard morphological operators of dilation, erosion and set difference respectively. The intermediate image, H, is finally thresholded by a value T to result in the Top Hat image by
Th=1 if H≧T;
Th=0 if H<T
where Th is the Top Hat image. The Jackway Top Hat transform is described in detail in Electronics Letters, Vol 36, No 14, pg 1194–5. Other possible Top Hat transforms are referred to in the Electronics Letters paper.
Referring to FIG. 3 , the down-sampled gray-scale image A′ (bottom image in FIG. 3 ), is treated as a landscape (top image in FIG. 3 ), in which higher altitude levels represent darker pixels in the gray-scale image and lower altitude levels represent lighter pixels.
The Top Hat transform works by passing a three-dimensional shape, that of a top hat shape, over a three-dimensional landscape. The rim of the top hat is placed at every location of the landscape such that it rests on the highest part of the landscape underneath the top hat. The top hat maintains the same attitude, i.e. it cannot rest on two different levels and be at a slant. If a part of the image landscape penetrates the top of the top hat, that part of the image becomes labelled in the transformed image. If the top hat is not penetrated by a particular part of the image, that part is not labelled in the transformed image. This is illustrated in two dimensions in FIG. 4 .
The result T′ of the Top Hat transform of the down-sampled image A′ of FIG. 3 , is shown in FIG. 5 . Parts of the landscape that penetrate the top hat are shown in white and those that did not are shown in black. A comparison of FIG. 5 with the input cell image in FIG. 2 shows that the location of some of the white areas of the Top Hat image correspond to the locations of objects-of-interest.
The next step in the scene segmentation process is to remove the white regions shown in FIG. 5 that do not correspond to objects-of-interest. This is achieved by eroding the image of FIG. 5 . Morphological erosion is a preferred technique of removing pixels from regions of an image to make the regions uniformly smaller. A suitable process for binary morphological erosion by a factor K1 is shown in the following pseudo-code:
Block: “Binary Erosion by factor K1” |
Inputs: Image T′, factor K1 | |
Output: Image I′ | |
Algorithm: | |
NewImage = Image T′ | |
FOR steps = 1 TO K1 |
OldImage = NewImage | |
FOR each Pixel in NewImage |
Pixel NewImageX,Y = |
MIN(OldImageX−1,Y−1, OldImageX−1,Y, | |
OldImageX−1,Y+1, |
OldImageX,Y−1, OldImageX,Y, OldImageX,Y+1, | |
OldImageX+1,Y−1, OldImageX+1,Y, | |
OldImageX+1,Y+1) |
ENDFOR |
ENDFOR | ||
RETURN Image I′ = NewImage | ||
The result of eroding the image of FIG. 5 is the image I′ shown in FIG. 6 . It can be seen that some of the small white regions of FIG. 5 have been removed completely, whilst the larger white regions have been made smaller. At this stage in the process, the aim is that every white region in FIG. 6 corresponds to a point within an object-of-interest in the input image of FIG. 2 .
Using the image in FIG. 6 , it is potentially possible to generate regions-of-interest by drawing a box of fixed size around each white region in FIG. 6 . However, the method of the present invention uses a Watershed transform to obtain a better estimate of the regions-of-interest.
The Watershed transform requires two inputs. It requires a point within each object-of-interest, an inner marker, and a closed boundary outside each object-of-interest, an outer marker. The white regions in FIG. 6 serve as a good inner marker and the generation of a good outer marker, according to the present invention, is described in the following.
The image I′ of FIG. 6 is used as a starting point for the generation of an outer marker. An operation known as binary reconstruction is used to re-grow the white regions T′ of FIG. 5 from the white regions of I′ of FIG. 6 . However, only the white regions in FIG. 6 are used because these correspond to the objects-of-interest. The following pseudo-code demonstrates a suitable binary reconstruction method:
Block: “Binary Reconstruction” |
Inputs: Image T′, Marker Image I′ | |
Output: Image R′ | |
Algorithm: | |
FOR each Object in Image T′ |
IF AND(ObjectK, Marker I′) ≠ 0 |
Copy ObjectK to Image R′ |
ENDIF |
RETURN Image R′ | ||
The result of this operation is the image R′ shown in FIG. 7 . The reconstructed image R′ of FIG. 7 is then dilated by a factor K2 according to the following pseudo-code:
Block: “Binary Dilation by factor K2” |
Inputs: Image R′, factor K2 | |
Output: Image M′ | |
Algorithm: | |
NewImage = Image T′ | |
FOR steps = 1 TO K2 |
OldImage = NewImage | |
FOR each Pixel in NewImage |
Pixel NewImageX,Y = |
MAX(OldImageX−1,Y−1, OldImageX−1,Y, OldImageX−1,Y+1, |
OldImageX,Y−1, OldImageX,Y, OldImageX,Y+1, | |
OldImageX+1,Y−1, OldImageX+1,Y, OldImageX+1,Y+1) |
ENDFOR |
ENDFOR | ||
RETURN Image M′ = NewImage | ||
Morphological dilation is a preferred technique to produce the opposite effect of morphological erosion and, in this case, adds pixels to the white regions of FIG. 7 . The result M′ of dilation is shown in FIG. 8 .
Although the dilation step could occur before binary reconstruction, it is computationally wasteful to do so, since additional calculations are made that are then discarded.
The final output from the Watershed transform must be at the original input image size and therefore the transform must be performed at the original input image size. Thus, the inputs for the transform are required to be at the original input image size. Consequently, the dilated image M′ shown in FIG. 8 and the inner marker image I′ shown in FIG. 6 are up-sampled to the original image size. The factor by which the images are up-sampled, N, is the same as the factor by which the original input image was down-sampled to reduce the original image. The inner marker image I is the result of up-sampling I′, and is shown in FIG. 9 . The result of up-sampling M′ is the image M, which is shown in FIG. 10 .
The following pseudo-code exemplifies up-sampling of the image I′ to the image I. The same algorithm may be applied to up-sample M′ to M.
Block: “Up Sample by factor N” |
Inputs: Image I′, factor N | |
Output: Image I | |
Algorithm: | |
NewImage = Image I′ | |
FOR steps = 1 TO N |
OldImage = NewImage | |
NewImage = size(2*X_size(OldImage), 2*Y_size | |
(OldImage)) | |
FOR each Pixel in OldImage |
Pixel NewImage2X,2Y = OldImageX,Y | |
Pixel NewImage2X,2Y+1 = OldImageX,Y | |
Pixel NewImage2X+1,2Y = OldImageX,Y | |
Pixel NewImage2X+1,2Y+1 OldImageX,Y |
ENDFOR |
ENDFOR | ||
RETURN Image I = NewImage | ||
The outer marker required for the Watershed transform may be generated from the up-sampled dilated image by determining perimeters of the up-sampled white regions. The perimeters are shown in the image O in FIG. 11 .
One way of performing perimeterisation is by looking at each black background pixel in the input image and if any of it's neighbours are white then the corresponding pixel in the perimeter image is turned to white. All other cases result in a black pixel in the perimeter image. The process is represented in the following pseudo-code:
Block: “Perimeter” |
Input: Image M | |
Output: Image O | |
Algorithm: | |
FOR each Pixel in Image M |
IF Pixel MX,Y = 0 |
IF any Neighbour(Pixel MX,Y) ≠ 0 |
SET Pixel OX,Y = 1 |
ENDIF |
ENDIF |
RETURN Image O | ||
The result of the perimeterisation algorithm is the image O shown in FIG. 11 . The positioning of the outer markers of FIG. 11 and the inner markers of FIG. 9 , relative to the original input image of FIG. 2 , is shown in FIG. 12 .
The next step in the process is to apply a Watershed algorithm such as is described by Beucher and Meyer in chapter 12, pages 433–482 of “Mathematical Morphology in Image Processing”, published in New York by Marcel Dekker in 1993. The result of the Watershed transform using the generated inner and outer markers as inputs, is image W which is shown in FIG. 13 . It can be seen that each object-of-interest has been delineated.
The final step in the process is to extract the required regions-of-interest, ROI. A bounding box is placed around each object-of-interest and the resulting final set of region-of-interest images, ROI, is shown in FIG. 14 . The following algorithm is suitable for the process:
Block: “Object Extraction” |
Inputs: Image W, Border Size M | |
Outputs: Regions of Interest Image ROI | |
Algorithm: | |
Xmin=1024 | |
Xmax=0 | |
Ymin=1024 | |
Ymax=0 | |
FOR each Object in Image W /* if any */ |
FOR each PixelX,Y in ObjectK |
Xmin = MIN(X, Xmin) | |
Xmax = MAX(X, Xmax) | |
Ymin = MIN(Y, Ymin) | |
Ymax = MAX(Y, Ymax) |
ENDFOR | |
Xmin = Xmin − M /* add M Pixel border all round*/ | |
Xmax = Xmax + M | |
Ymin = Ymin − M | |
Ymax = Ymax + M | |
SET ROI = getImageRegion(W, Xmin, Xmax, Ymin, Ymax) | |
OUTPUT Image ROI |
ENDFOR | ||
These images may be used for further segmentation processing, such as for the second segmentation sub-task mentioned above.
Throughout the specification the aim has been to describe the preferred embodiment of the present invention without limiting the invention to any one embodiment or specific collection of features. Persons skilled in the relevant art may realize variations from the specific embodiments that will nonetheless fall within the scope of the invention.
Claims (9)
1. A method of scene segmentation comprising the steps of:
(i) transforming an input image by a Top Hat operator;
(ii) conducting erosion on said Top Hat transformed image to produce an inner marker image;
(iii) eliminating objects less than an object-of-interest size by reconstructing said Top Hat transformed image with said inner marker image to produce a reconstructed image;
(iv) conducting dilation on said reconstructed image to produce an outer marker image;
(v) determining a perimeter of said outer marker image; and
(vi) transforming said input image using said inner marker image and said perimeterised outer marker image by a Watershed operator to produce a segmented image of objects.
2. The method of claim 1 further comprising the step of delineating a region-of-interest around each object to produce region-of-interest images.
3. The method of claim 1 further comprising the preliminary step of down-sampling the input image by a factor to produce a smaller image.
4. The method of claim 1 further comprising the step of up-sampling the eroded and dilated images by a factor.
5. The method of claim 1 further comprising the steps of down-sampling the input image by a factor N to produce a smaller image and up-sampling the eroded and dilated images by a factor of M.
6. The method of claim 5 wherein the factors N and M are the same such that the inner and outer marker images are the same size as that of the input image.
7. The method of claim 1 wherein the Top Hat operator is of the form:
H=f−((f⊕(B do \B di))⊖B di),
H=f−((f⊕(B do \B di))⊖B di),
where f is the original image, Bdo and Bdi are outer and inner structuring elements respectively, ⊕, ⊖ and \ are standard morphological operators of dilation, erosion and set difference respectively.
8. The method of claim 1 wherein said erosion is morphological erosion.
9. The method of claim 1 wherein said dilation is morphological dilation.
Applications Claiming Priority (2)
Application Number | Priority Date | Filing Date | Title |
---|---|---|---|
AUPQ8492A AUPQ849200A0 (en) | 2000-06-30 | 2000-06-30 | Unsupervised scene segmentation |
PCT/AU2001/000787 WO2002003331A1 (en) | 2000-06-30 | 2001-06-28 | Unsupervised scene segmentation |
Publications (2)
Publication Number | Publication Date |
---|---|
US20030169946A1 US20030169946A1 (en) | 2003-09-11 |
US7142732B2 true US7142732B2 (en) | 2006-11-28 |
Family
ID=3822559
Family Applications (1)
Application Number | Title | Priority Date | Filing Date |
---|---|---|---|
US10/312,508 Expired - Fee Related US7142732B2 (en) | 2000-06-30 | 2001-06-28 | Unsupervised scene segmentation |
Country Status (7)
Country | Link |
---|---|
US (1) | US7142732B2 (en) |
EP (1) | EP1309944B1 (en) |
JP (1) | JP2004502262A (en) |
AU (3) | AUPQ849200A0 (en) |
CA (1) | CA2414579A1 (en) |
DE (1) | DE60140810D1 (en) |
WO (1) | WO2002003331A1 (en) |
Cited By (3)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
US20100284618A1 (en) * | 2007-12-31 | 2010-11-11 | Dimitrios Ioannou | Method and system for identifying an object |
CN103268476A (en) * | 2013-05-14 | 2013-08-28 | 中国科学院自动化研究所 | Target monitoring method of remote sensing image |
US20180182100A1 (en) * | 2016-12-23 | 2018-06-28 | Bio-Rad Laboratories, Inc. | Reduction of background signal in blot images |
Families Citing this family (11)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
AUPR647801A0 (en) | 2001-07-19 | 2001-08-09 | Cea Technologies Inc. | Chromatin segmentation |
US6985612B2 (en) * | 2001-10-05 | 2006-01-10 | Mevis - Centrum Fur Medizinische Diagnosesysteme Und Visualisierung Gmbh | Computer system and a method for segmentation of a digital image |
GB0410499D0 (en) * | 2004-05-11 | 2004-06-16 | Fairfield Imaging Ltd | Method and apparatus for use in the image analysis of biological specimens |
JP2006068373A (en) * | 2004-09-03 | 2006-03-16 | Fuji Photo Film Co Ltd | Mammilla detector and program thereof |
US8206301B2 (en) * | 2005-02-01 | 2012-06-26 | Fujifilm Corporation | Ultrasonic imaging apparatus and ultrasonic image processing apparatus, method and program |
US7817841B2 (en) * | 2005-11-12 | 2010-10-19 | General Electric Company | Time-lapse cell cycle analysis of unstained nuclei |
US8189943B2 (en) * | 2009-03-17 | 2012-05-29 | Mitsubishi Electric Research Laboratories, Inc. | Method for up-sampling depth images |
KR101401734B1 (en) | 2010-09-17 | 2014-05-30 | 한국전자통신연구원 | Apparatus and method for detecting examination tissue using ultrasound signal |
KR102264591B1 (en) * | 2015-02-27 | 2021-06-15 | 삼성전자주식회사 | Image Processing Method and Electronic Device supporting the same |
CN107971240B (en) * | 2017-11-21 | 2019-06-04 | 合肥工业大学 | A kind of color selection method of electrical core of power battery breakdown products copper aluminium foil particle |
CN108470173B (en) * | 2018-04-17 | 2021-06-15 | 北京矿冶科技集团有限公司 | Ore particle segmentation method |
Citations (8)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
US5257182A (en) * | 1991-01-29 | 1993-10-26 | Neuromedical Systems, Inc. | Morphological classification system and method |
US5768407A (en) * | 1993-06-11 | 1998-06-16 | Ortho Diagnostic Systems, Inc. | Method and system for classifying agglutination reactions |
US5850464A (en) * | 1996-01-16 | 1998-12-15 | Erim International, Inc. | Method of extracting axon fibers and clusters |
US5892841A (en) * | 1995-10-18 | 1999-04-06 | Societe Nationale D'etude Et De Construction De Moteurs D'aviation "Snecma" | Process for automatic detection of the assessable areas in images of mechanical components |
US6195659B1 (en) * | 1998-07-14 | 2001-02-27 | Trw Inc. | Method and apparatus for morphological clustering having multiple dilation and erosion of switchable grid data cells |
US6244764B1 (en) * | 2000-01-21 | 2001-06-12 | Robotic Vision Systems, Inc. | Method for data matrix print quality verification |
US6363161B2 (en) * | 1998-10-23 | 2002-03-26 | Facet Technology Corp. | System for automatically generating database of objects of interest by analysis of images recorded by moving vehicle |
US6400831B2 (en) * | 1998-04-02 | 2002-06-04 | Microsoft Corporation | Semantic video object segmentation and tracking |
Family Cites Families (1)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
FR2755148B1 (en) * | 1996-10-31 | 1999-01-29 | Calvo Fabien | METHOD FOR HIGHLIGHTING THE CONDITION OF A MALIGNANT CELL AND TREATMENT METHOD |
-
2000
- 2000-06-30 AU AUPQ8492A patent/AUPQ849200A0/en not_active Abandoned
-
2001
- 2001-06-28 US US10/312,508 patent/US7142732B2/en not_active Expired - Fee Related
- 2001-06-28 DE DE60140810T patent/DE60140810D1/en not_active Expired - Fee Related
- 2001-06-28 AU AU6716001A patent/AU6716001A/en active Pending
- 2001-06-28 WO PCT/AU2001/000787 patent/WO2002003331A1/en active IP Right Grant
- 2001-06-28 AU AU2001267160A patent/AU2001267160B2/en not_active Ceased
- 2001-06-28 CA CA002414579A patent/CA2414579A1/en not_active Abandoned
- 2001-06-28 JP JP2002507322A patent/JP2004502262A/en active Pending
- 2001-06-28 EP EP01944759A patent/EP1309944B1/en not_active Expired - Lifetime
Patent Citations (10)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
US5257182A (en) * | 1991-01-29 | 1993-10-26 | Neuromedical Systems, Inc. | Morphological classification system and method |
US5257182B1 (en) * | 1991-01-29 | 1996-05-07 | Neuromedical Systems Inc | Morphological classification system and method |
US5768407A (en) * | 1993-06-11 | 1998-06-16 | Ortho Diagnostic Systems, Inc. | Method and system for classifying agglutination reactions |
US5892841A (en) * | 1995-10-18 | 1999-04-06 | Societe Nationale D'etude Et De Construction De Moteurs D'aviation "Snecma" | Process for automatic detection of the assessable areas in images of mechanical components |
US5850464A (en) * | 1996-01-16 | 1998-12-15 | Erim International, Inc. | Method of extracting axon fibers and clusters |
US6400831B2 (en) * | 1998-04-02 | 2002-06-04 | Microsoft Corporation | Semantic video object segmentation and tracking |
US6195659B1 (en) * | 1998-07-14 | 2001-02-27 | Trw Inc. | Method and apparatus for morphological clustering having multiple dilation and erosion of switchable grid data cells |
US6363161B2 (en) * | 1998-10-23 | 2002-03-26 | Facet Technology Corp. | System for automatically generating database of objects of interest by analysis of images recorded by moving vehicle |
US6625315B2 (en) * | 1998-10-23 | 2003-09-23 | Facet Technology Corp. | Method and apparatus for identifying objects depicted in a videostream |
US6244764B1 (en) * | 2000-01-21 | 2001-06-12 | Robotic Vision Systems, Inc. | Method for data matrix print quality verification |
Non-Patent Citations (2)
Title |
---|
P.T. Jackway, Improved Morphological Top-Hat, Electronics Letters, vol. 36(4), Jul. 2000, pp. 1194-1195, Institution of Electrical Engineers. |
Qian et al., Automatic Extraction of Coronary Artery Tree on Coronary Angiograms by Morphological Operators, Computers in Cardiology, vol. 25, 1998, pp. 765-768, IEEE. |
Cited By (5)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
US20100284618A1 (en) * | 2007-12-31 | 2010-11-11 | Dimitrios Ioannou | Method and system for identifying an object |
CN103268476A (en) * | 2013-05-14 | 2013-08-28 | 中国科学院自动化研究所 | Target monitoring method of remote sensing image |
CN103268476B (en) * | 2013-05-14 | 2016-07-06 | 中国科学院自动化研究所 | A kind of Remote Sensing Target monitoring method |
US20180182100A1 (en) * | 2016-12-23 | 2018-06-28 | Bio-Rad Laboratories, Inc. | Reduction of background signal in blot images |
US10846852B2 (en) * | 2016-12-23 | 2020-11-24 | Bio-Rad Laboratories, Inc. | Reduction of background signal in blot images |
Also Published As
Publication number | Publication date |
---|---|
AU2001267160B2 (en) | 2005-10-13 |
AU6716001A (en) | 2002-01-14 |
CA2414579A1 (en) | 2002-01-10 |
JP2004502262A (en) | 2004-01-22 |
US20030169946A1 (en) | 2003-09-11 |
AUPQ849200A0 (en) | 2000-07-27 |
DE60140810D1 (en) | 2010-01-28 |
WO2002003331A1 (en) | 2002-01-10 |
EP1309944A1 (en) | 2003-05-14 |
EP1309944A4 (en) | 2009-01-21 |
EP1309944B1 (en) | 2009-12-16 |
Similar Documents
Publication | Publication Date | Title |
---|---|---|
EP1304651B1 (en) | Background-based image segmentation | |
JP6100744B2 (en) | Color document image segmentation and binarization using automatic restoration | |
EP0664038B1 (en) | Method for identifying objects using data processing techniques | |
US7142732B2 (en) | Unsupervised scene segmentation | |
JPH1023452A (en) | Picture extracting device and its method | |
Géraud et al. | Color image segmentation based on automatic morphological clustering | |
AU2001267160A1 (en) | Unsupervised scene segmentation | |
EP2034436A1 (en) | Method for non-photorealistic rendering | |
GB2466818A (en) | Cell image segmentation using binary threshold and greyscale image processing | |
Chen et al. | Single depth image super-resolution using convolutional neural networks | |
Falah et al. | A region-region and region-edge cooperative approach of image segmentation | |
CN112884662A (en) | Three-dimensional digital map image processing method based on aerial image of aircraft | |
Sarker et al. | Morphological based technique for image segmentation | |
Wilkinson | Automated and manual segmentation techniques in image analysis of microbes | |
Zambanini et al. | Robust automatic segmentation of ancient coins | |
CN114529518A (en) | Image pyramid and NLM-based image enhancement method for cryoelectron microscope | |
Chehdi et al. | A blind system to identify and filter degradations affecting an image | |
Mtimet et al. | A layer-based segmentation method for compound images | |
Gupta et al. | Fully automated image de-fencing using conditional generative adversarial networks | |
CN115984316B (en) | Industrial image edge extraction method and device for complex environment | |
Das et al. | Enhancement of image edge sharpness and acutance | |
Joshi et al. | Lung Cancer Detection Using Image Processing | |
CN116957976A (en) | Target image enhancement method, target image enhancement device, and storage medium | |
CN115861482A (en) | Background filling method, device and equipment | |
Rai et al. | Segmentation Based Denoising of Color Images using Morphological Operation |
Legal Events
Date | Code | Title | Description |
---|---|---|---|
AS | Assignment |
Owner name: CO-OPERATIVE RESEARCH CENTRE FOR SENSOR SIGNAL AND Free format text: ASSIGNMENT OF ASSIGNORS INTEREST;ASSIGNORS:BAMFORD, PASCAL;JACKWAY, PAUL;REEL/FRAME:014135/0839 Effective date: 20010630 |
|
FPAY | Fee payment |
Year of fee payment: 4 |
|
REMI | Maintenance fee reminder mailed | ||
LAPS | Lapse for failure to pay maintenance fees | ||
STCH | Information on status: patent discontinuation |
Free format text: PATENT EXPIRED DUE TO NONPAYMENT OF MAINTENANCE FEES UNDER 37 CFR 1.362 |
|
FP | Lapsed due to failure to pay maintenance fee |
Effective date: 20141128 |